Heterogeneous Multi-Agent LLM Framework for Automated AI Governance and Compliance Assessment under the UAE Information Assurance Standard V2
Alblooshi, Abdulaziz Mousa Ibrahim
Alblooshi, Abdulaziz Mousa Ibrahim
Citations
Altmetric:
Supervisor
Department
Computer Science
Embargo End Date
Type
Thesis
Date
2026
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
The pace of artificial intelligence being embraced by government organisations, critical infrastructure, and commercial services has exceeded the maturity of governance frame- works, leading to a lack of uniform risk assessment, unprovable claims to compliance, and limited traceability of audit. In this thesis, a heterogeneous multi-agent large language model (LLM) architecture for automated AI governance and compliance evaluation within the UAE Information Assurance Standard V2 (2025) is presented. Three role-specialised agents (Auditor, CISO, and CEO) independently rate AI use cases based on structured control families, and disagreements are resolved through a conservative majority-voting consensus mechanism. Tested on five representative scenarios of AI deployment in the UAE public sector with ten experimental settings, the framework achieves 90% Risk Tier Accuracy, which is 20–40 percentage points better than single-LLM solutions (50–70%) and 5–10 percentage points better than same-model multi-agent conditions (80–85%). Findings are consistent across all three role-to-model rotation settings, demonstrating that the accu- racy improvement is due to model heterogeneity and not particular role assignment. The framework does not require manual post-processing, allowing it to scale and provide an objective way of assessing AI compliance through fully traceable, audit-ready governance reports.
Citation
Alblooshi, Abdulaziz Mousa Ibrahim, "Heterogeneous Multi-Agent LLM Framework for Automated AI Governance and Compliance Assessment under the UAE Information Assurance Standard V2," M.S. Thesis, Computer Science, MBZUAI, 2026.
Source
Conference
Keywords
Multi-Agent Systems, AI Governance, Compliance Automation, Large Language Models, UAE Information Assurance Standard, Risk Assessment